slingshot to understand progression of the disease

[FADHLyemen]

Hi @kstreet13 sorry, I do not see a discussion panel here so this is not an issue.

I have two questions: do you plan to have trajectories workshop for this year?

second question: I have three scRNA-seq datasets, early, mid and late disease. Can I integrate them to understand the progression of the disease? do you have a real example for this scenario? If I identify genes differentiated between progression stages among predicted trajectories, can I use the slingshot to simulate how knocking these genes or one gene will cut or break or stop the trajectory from progressing from early to late-stage?

[kstreet13]

Hi @FADHLyemen,

I don't think we will have another workshop this year, unfortunately (at least not this summer, as far as I know).

As for integrating the different timepoints, I know of a few ways to do this. I've generally found good results with fastMNN from the batchelor package, which represents all the cells in a shared dimensionality reduction, similar to PCA. We have also used the Seurat anchor-based integration workflow for some of our workshops with pretty reasonable results. And many people use Harmony for this sort of integration, though I don't have as much experience with that tool.

And unfortunately, I don't think it's possible to use Slingshot + tradeSeq to predict the effects of a gene knock out experiment. They may be useful tools for analyzing a particular experiment and identifying genes of interest, but they generally don't account for interdependencies between genes. Given my limited understanding of biology, I think that question sounds really interesting but also extremely challenging.

Best, Kelly

[FADHLyemen]

Thank you @kstreet13 I will use tradseq to identify genes that differ between early and late among slingshot inferred trajectory. Then I will replace these values for these genes with values to make early and late comparable. I will reintegrate the data using seurat then use the slingshot to infere the trajectory. if the trajectory does not move from early to late that means, these genes are responsible for disease progression. is this look reasonable to you?

[kstreet13]

No, I don't think that will provide meaningful results. In a biological system, changing one gene can have downstream effects on many others. But the procedure you are proposing would only change the data for one gene, leaving all others the same. You might get slightly different results when you re-run the whole pipeline with this altered dataset, but I would not put any faith in them as predictions.

[FADHLyemen]

Thank you, is it reasonable to compare trajectories for perturbed and not perturbed datasets?

[kstreet13]

Yes, comparing trajectories across conditions is the aim of our workflow and is addressed by the condiments package.